Taste Over Torque: Why Craft Still Matters in the Age of Instant AI

AI doesn’t care.

It will gleefully ship a million “good-enough” versions before you’ve finished picking a font. It never loses sleep over kerning, never blushes at a clunky product demo, never feels that microscopic wince when a spinner lingers a half-second too long. That indifference is both its superpower and its fatal flaw.

From Scarcity to Saturation

Investor Anu Atluru framed it perfectly:

“In a world of scarcity, we treasure tools. In a world of abundance, we treasure taste.”

Speed and execution used to be a moat. Today—thanks to GPT-powered code, one-click design suites, and infinite outbound—speed is table stakes. Your competitor can push ten deploys before lunch; so can you.

What they can’t mass-produce is judgment.

Spotify’s engineered shuffle that sidesteps repetition, the tiny haptic nudge in an iPhone’s mute switch, the impossible-to-miss clarity of Figma’s onboarding flow—these are outcomes of taste, not throughput. As Sarah Guo and Felix Haas have argued, that human filter is the new defensibility.

The 90-Percent Trap

AI now carries us to 90 percent in 30 minutes. The temptation is obvious: ship it, tweet the launch thread, call it a day. But 90 percent is usually the uncanny valley—polite, competent, forgettable. The final 10 percent is where affinity lives.

Taste hurts:

the feature you cut because it dilutes the story,
the prospect you disqualify because the fit is wrong,
the headline you rewrite at 1 a.m. because a single verb felt lazy.

No model will feel that sting for you. That’s the job.

How We Build at Compelling

I spent thirteen years in advertising zooming to 400 percent just to nudge a pixel. Founding Compelling taught me that startup clocks run on dog years; perfectionism feels criminally expensive. Yet our north star hasn’t changed:

Automation should amplify human taste, not replace it.

That’s why every Compelling release has two checkboxes:

Does it save reps measurable time?
Does it feel like a tool designed by people who care?

If either answer is “no,” we keep grinding—even if the AI models are already patting themselves on the back.

The Moat You Can’t Fine-Tune

Your prospects can smell a robot-assembled product. They can also feel the difference when a team sweats the final 0.1 percent. In a world of infinite versions, the one with taste wins. AI doesn’t care. We do—and so will our customers.

So go ahead, let the machines sprint. Just make sure that when they hand off the baton, there’s a human waiting to finish the race with style.

Article by